A Pipeline to Construct a Real-World Question and GeoSPARQL Query Dataset for Geo-Analytical Question Answering over OpenStreetMap Data

Abstract

An immense pull of data led to the growth of the QA (Question Answering) system. With the growth of comprehensive QA systems, KBQA (QA over Knowledge Base) demonstrates an effective way of answering questions based on knowledge sources. GeoQA (Geographic Question Answering), in turn, still lacks its research and advancement, even with the rapid increase in geospatial data. There is only one full set of question and SPARQL query pair datasets specific to GeoQA, limiting its potential to become a comprehensive GeoQA system. In order to build a comprehensive GeoQA system, this paper proposes a pipeline to construct a real-world question and GeoSPARQL query pair datasets for geo-analytical questions over OSM (OpenStreetMap) data. Through utilizing the real-world MS MARCO (Machine Reading Comprehensive) question dataset, we classify them into geometry and operation combinations and generate a geo-analytical workflow for further query generation. We evaluate a detailed comparison with the existing Geospatial Gold Standard 201 question, query pair dataset to give context as well.